PROPRIEDADES ÓPTICAS DE FOLHAS DE MILHO EM FUNÇÃO DE DIFERENTES TEORES DE PIGMENTOS E ESTIMATIVA NÃO DESTRUTIVA DO TEOR FOLIAR DE FÓSFORO E DA PRODUTIVIDADE

Abstract

Technological advances have allowed studies and the search for new solutions in agribusiness, producing reliable results with lower costs and delivery time. Remote sensing can extract information from a specific target without maintaining physical contact. Among the equipment capable of sensing the radiation reflected by plants, the spectroradiometers stand out, due its ability to detect small differences in the biophysical and/or biochemical characteristics of the plants, being able to generate accurate and useful information for agriculture. The aim of the study was to characterize the spectral behavior of maize as a function of different leaf anthocyanin levels, generating regression models, by partial least squares (PLSR) and support vectors(SVR, to predict leaf phosphorus content, also creating a regression model, by minimum squares, to predict maize yield. The study was carried out for two years (2018 and 2019) in the city of Campo Mourão, Paraná State. The experimental design was a randomized complete blocks, testing (0, 65, 200, 260 and 330 kg ha-1 of P2O5) with five replications. The evaluations were performed in four phenological stages (V8, VT, R3 and R5). The readings of the radiation reflected by the leaves were taken using the ASD Fieldspec 3 Jr. spectroradiometer. Unscrambler® software was used in all statistical analysis. The best period for the prediction of leaf phosphorus content, using the PLSR model, was at R3, and for the SVR model at VT. The success rate of interpretation classes for leaf phosphorus content with the PLSR model was 86% and 93 % using the SVR model. As the fertilizer rate increased, the levels of flavonoids and anthocyanins were lower and those of chlorophyll and carotenoids were higher. There was a spectral behavior difference in the 400-800 nm range due to the treatments, mainly between 500-600 nm, with a reflectance peak at 550 nm. In both first derivative, the reflectance peak shifted to the left (close to 520 nm). The PLSR model, generated from data of the first derivative, obtained the best statistical parameters to predict anthocyanins. The best period to predict maize yield was at VT, and the statistical parameters observed in the external validation were R2 0.60; RMSE 619.32; BIAS 117.87 and SE 629.36.